Fix Phobos dependencies on autodecoding

[Walter Bright]

I ran into that as well with the 3 PRs I did:

fix array(String) to work with no autodecode
https://github.com/dlang/phobos/pull/7133

fix assocArray() unittests for no autodecode
https://github.com/dlang/phobos/pull/7134

fix unittests for array.join() for no autodecode
https://github.com/dlang/phobos/pull/7135

More specifically, the ElementType template returns a dchar for an autodecodable 
string, and char/wchar for a non-autodecodable string. I suspect that most 
people are not aware of this, and code that uses ElementType may already be 
subtly broken.

Note that the documentation for ElementType is also wrong,

   https://github.com/dlang/phobos/pull/

because isNarrowString is NOT THE SAME THING as an autodecoding string! The 
difference is isNarrowString excludes stringish aggregates and enums with a 
string base type, while autodecoding types include them.

Does this confuse anyone? It confuses me. I can never remember which is which.

Autodecoding is not only a conceptual mistake, the way it is implemented is a 
buggy, confusing disaster. (isNarrowString is often incorrectly used instead 
isAutodecodableString in Phobos.)

I think the only solution is to "rip the band-aid off" and have ElementType give 
the code unit type when autodecoding is disabled.